U.S. Pulls Plug On Top AI

The world’s most advanced public AI just went dark overnight, not because it broke the rules—but because Washington says it might.

Story Snapshot

The U.S. government ordered Anthropic to cut off its top AI models, Fable 5 and Mythos 5, worldwide over “national security” fears.
The directive targets foreign nationals, but Anthropic says the only way to comply was to shut the models off for everyone.^[2]^[3]
Officials pointed to a narrow “jailbreak” claim, which Anthropic disputes as minor and no stronger than what rival models already do.^[2]^[5]
The shutdown shows how new export-control powers now reach into cloud software, not just chips and weapons.

Government Order Pulls the Plug on Fable 5

Late on a Friday evening, just three days after launch, Anthropic’s most powerful models, Fable 5 and Mythos 5, were taken offline for every user on earth.^[3] The company says the U.S. government, citing national security, sent an export-control directive at 5:21 p.m. Eastern time that banned access for any foreign national, even those living and working inside the United States.^[2]^[3] Because the service runs in a shared cloud, Anthropic says it could not reliably filter by nationality, so it had to shut the models off for everyone.^[2]^[4]

Anthropic’s public statement is blunt: “The U.S. government… has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national.”^[2] The company adds, “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.”^[2] Importantly, Anthropic says all of its other models remain available, so this is not a systemwide outage but a targeted shutdown of the most capable systems only.^[2]^[3]

Thin “Jailbreak” Evidence, Thick Regulatory Muscle

The government’s concern, according to Anthropic, is a reported “jailbreak” method that could bypass some of Fable 5’s safety rules.^[2]^[5] Anthropic says it saw the demo and concluded it only revealed a small number of minor software vulnerabilities that were already known and that did not unlock broad cyberattack powers.^[2]^[5] The company also says the level of capability shown in the report is already widely available in other top models, including OpenAI’s GPT‑5.5, which remain online.^[3]^[5]

So far, Anthropic says it has not even received a written technical report of a harmful exploit, only verbal claims of a narrow, non‑universal jailbreak that involves asking the model to read a specific codebase and identify bugs.^[2]^[3] The company argues that its strongest protections come from separate safety systems that continue to block dangerous outputs even if the core model is prodded past its first refusal.^[5] In other words, Washington used maximum export‑control force based on a disputed, narrow bug that the company believes does not justify this kind of sweeping action.^[2]^[5]

From Chips and Missiles to Cloud AIs in Your Browser

This shutdown is not happening in a vacuum. Since 2022, the U.S. government has steadily expanded export controls that once focused on missiles and high‑end semiconductors into the world of artificial intelligence models and cloud access. A 2026 House bill called the Remote Access Security Act aims to write into law that export‑control rules apply even when a foreign person only “uses” advanced computing over the internet, with no hardware ever crossing a border. That is the legal logic behind treating log‑ins to Fable 5 like physical exports.

Policy analysts warn that “catch‑all” rules under the Export Administration Regulations already let officials block any U.S.‑origin software if they think it might help foreign militaries or intelligence services. Recent enforcement cases show the Department of Justice and the Commerce Department’s Bureau of Industry and Security going after chip smugglers, cloud operators, and even finance middlemen tied to China’s high‑end computing sector. Now, with Fable 5 and Mythos 5, that same machinery is being used on consumer‑facing AI tools accessed by ordinary developers, students, and small businesses around the world.^[2]^[3]

What This Means for Free Speech, Innovation, and American Power

Supporters of strong export controls say Washington cannot ignore the risk that advanced models might be turned into cyberweapons or used to supercharge foreign militaries. They argue that restricting access for foreign nationals is a reasonable step when the United States believes a model can find deep software flaws at machine speed. But Anthropic points out that Mythos 5 has also been used by U.S. authorities to locate and fix long‑hidden security gaps, meaning the same tool can make American systems safer too. Sweeping shutdowns could weaken defenders along with attackers.

On Friday night, the tech world saw something completely unprecedented: the global, sudden shutdown of Anthropic’s Claude Fable 5 and Mythos 5.

Not throttled. Not fenced. Switched off globally by an undisclosed government export-control directive—just three days after launch.…

— ARAGS Inc. (@ARAGS_Inc) June 15, 2026

For everyday Americans, the bigger story is precedent. If a narrow, disputed jailbreak can trigger a worldwide shutdown of the most advanced public model, every new frontier system now lives under the shadow of sudden government switch‑offs.^[3]^[5] Anthropic itself warns that if this standard is applied across the board, it would “effectively halt” new model deployments from every major lab. That means fewer tools for small businesses, slower growth in next‑generation productivity, and more power in the hands of regulators and large, politically connected players who can navigate the rules.